v75 (Galaxy S24 / SM8650, soc_model 57): Llama-3.1-Nemotron-Nano-8B 8-shard W8 bundle — export 8/8 (runtime needs >8GB RAM)

Adds the Hexagon v75 flat bundle under `v75/` (Snapdragon 8 Gen 3 class).

- **Files:** nemotron-nano-8b.json, nemotronnano8b_decode_p0..p7_w8.bin (8-shard W8 decode, ~0.9G each), nemotronnano8b_lmhead_w8.bin, nemotronnano8b_embed_f16.bin, tokenizer.json (~8.1G total)
- **Precision:** 8-shard W8 decode + W8 lm-head + fp16 embed (large-LLM decode-split path, host-op llama_generate_sharded)
- **Compile:** QAIRT 2.47.0, dsp_arch v75, soc_model 57; export gate decode 8/8 match HF greedy
- **Device note:** all 8 decode shards load on v75, but the ~8G resident W8 weights exceed the **8GB** Galaxy S24 RAM at runtime (adb drop). Runs on higher-RAM v75 devices (v81 measured ~6GB RSS). Published for arch parity with v79/v81.

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 v81/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+v75/tokenizer.json filter=lfs diff=lfs merge=lfs -text

v75/nemotron-nano-8b.json

@@ -0,0 +1,67 @@
+{
+  "schema_version": 1,
+  "_comment": "Llama-3.1-Nemotron-Nano-8B-v1 (NVIDIA reasoning fine-tune of Llama-3.1-8B) on Hexagon v81 (SM8850 / soc_model 87). Standard dense Llama-3.1-8B (32 q heads -> materialized MHA + MAXCTX 512; llama3-scaled RoPE theta 500000 factor 8; untied lm-head). The ~7GB W8 decode exceeds the v81 cDSP per-context ceiling, so the 32 layers are split into 8 CHAINED decode parts (decode_parts=8, ~0.85GB W8 each) driven by the llama_generate_sharded host-op (decode-over-prompt seed, no batched prefill). Export gate: decode 8/8 match gold (forge). W8 device gate is greedy_tol.",
+  "model": {
+    "name": "nemotron-nano-8b",
+    "family": "llm",
+    "dsp_arch": "v75",
+    "tokenizer_pre": "llama3"
+  },
+  "params": {
+    "hidden": 4096,
+    "vocab": 128256,
+    "n_layers": 32,
+    "max_ctx": 512,
+    "kv_dim": 1024,
+    "head_dim": 128,
+    "rope_theta": 500000.0,
+    "eos_token_id": 128001
+  },
+  "artifacts": {
+    "contexts": {
+      "decode_p0": {
+        "bin": "nemotronnano8b_decode_p0_w8.bin"
+      },
+      "decode_p1": {
+        "bin": "nemotronnano8b_decode_p1_w8.bin"
+      },
+      "decode_p2": {
+        "bin": "nemotronnano8b_decode_p2_w8.bin"
+      },
+      "decode_p3": {
+        "bin": "nemotronnano8b_decode_p3_w8.bin"
+      },
+      "decode_p4": {
+        "bin": "nemotronnano8b_decode_p4_w8.bin"
+      },
+      "decode_p5": {
+        "bin": "nemotronnano8b_decode_p5_w8.bin"
+      },
+      "decode_p6": {
+        "bin": "nemotronnano8b_decode_p6_w8.bin"
+      },
+      "decode_p7": {
+        "bin": "nemotronnano8b_decode_p7_w8.bin"
+      },
+      "lmhead": {
+        "bin": "nemotronnano8b_lmhead_w8.bin"
+      }
+    },
+    "embed": "nemotronnano8b_embed_f16.bin",
+    "tokenizer": "tokenizer.json"
+  },
+  "plan": {
+    "steps": [
+      {
+        "host": "llama_generate_sharded",
+        "params": {
+          "decode_parts": "nemotronnano8b_decode_p0_w8,nemotronnano8b_decode_p1_w8,nemotronnano8b_decode_p2_w8,nemotronnano8b_decode_p3_w8,nemotronnano8b_decode_p4_w8,nemotronnano8b_decode_p5_w8,nemotronnano8b_decode_p6_w8,nemotronnano8b_decode_p7_w8",
+          "lmhead": "nemotronnano8b_lmhead_w8",
+          "max_new": 64,
+          "neg": -50000,
+          "rope_factor": 8.0
+        }
+      }
+    ]
+  }
+}

v75/nemotronnano8b_decode_p0_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b66e1eef4bc1903942bf09cb4823eddd9df57d20ea25d59244c4fa8d70eed76
+size 876392448

v75/nemotronnano8b_decode_p1_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b0a9aab17bbd5c27a433dc267011368279d6d14e1949c4c44af2d18e458e0df
+size 876392448

v75/nemotronnano8b_decode_p2_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca9a95042ae0e5f259174faca2cf14bd47b8169bd96c1e1d3771659773c6016b
+size 876392448

v75/nemotronnano8b_decode_p3_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a582d7bd4dbe8907ebe3e763329d8d3da7c5b58ea9ea1cc42d3262a4d871684a
+size 876392448

v75/nemotronnano8b_decode_p4_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb1463fcbdb1cedd3c7b7d181fc20e3c7d28541d00a7ff8d6d90d5b4ec23d3a0
+size 876392448

v75/nemotronnano8b_decode_p5_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5d1a6ea2a9fc086992b2e87f672c1d6bd8e3ae4cf5ec8078efcdeb69810bb5f
+size 876392448

v75/nemotronnano8b_decode_p6_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:543a315646fd0803230111d06f8de370bb5a6d7ab3efaf55b76fd4fb647852b8
+size 876392448

v75/nemotronnano8b_decode_p7_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ffbfd021e63a2bdd0cae5e904a47b19125c4a63f0fa3299d91ef7b13dbc2ac24
+size 876400640

v75/nemotronnano8b_embed_f16.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8443236c0685bf39fb998183ecd98d14e296cb6fee1e1c7841ec89d2c15fcfac
+size 1050673152

v75/nemotronnano8b_lmhead_w8.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:844a917ae15d026d9b418c04e9726a0619f90208aca7a5987d2b80841353cb8a
+size 527147008

v75/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b
+size 17209920